Graphite

✍️ Notes from the Editors

Good morning from SF! 🌉

We’re busy prepping for our founder dinner on Thursday. If you’re in town this week and want to say hi, reply and we’ll grab coffee.

There are some exciting announcements in this email… be sure to read through below to see what we’re planning in June. Hint: it involves some awesome events for a certain upcoming Tech Week.

- Arek and Ethan 🦄

Diamond-grade code reviews

Graphite streamlines code review workflows for engineering teams through stacked pull requests and Diamond, its AI‑powered reviewer. Stacked pull requests break larger changes into smaller sequential changes that are easier to review and merge. The platform plugs directly into common developer tools GitHub and VS Code, automatically keeping code up to date with teammates’ changes (known as rebasing). All the while, it summarizes its changes and suggests fixes with the goal of slashing review cycles and keeping pipelines unblocked.

🔗 Check it out: graphite.dev

🗽 Announcing: #NYTechWeek Takeover

Heading to New York Tech Week in June?

Unicorner is coming back to NYC for Tech Week with not one, not two... but FOUR #NYTechWeek events.

Reply to this email after applying, and we’ll get you front-of-the-line access as a Unicorner reader. 🦄 

• Tues. 6/3, 3pm: Fireside chat with Kalshi founder Tarek Mansour 📈

• Wed. 6/4, 3pm: Fireside chat with Bilt Rewards founder Ankur Jain 💳

• Wed. 6/4, 6pm: Emerging Fund Managers Omakase 🍣 

• Friday. 6/6, 8pm: Secret Society Afterparty with Flite 👀 

💰 Business Model

Graphite uses a SaaS business model with pricing based on the number of users, or seats, per month for its code review platform. ​Its AI code reviewer, Diamond, is available as an add-on for $15 per active contributor (known as committers on GitHub) per month with a Graphite subscription. Purchasing Diamond without a Graphite subscription costs $20 per active committer per month.

📈 Traction and Fundraising

• Raised $52 million Series B, led by Accel, with participation from Menlo Ventures and Anthropic via the Anthology Fund, Shopify Ventures, and Figma Ventures, as well as returning investors Andreessen Horowitz (a16z) and The General Partnership 

• Grown from a handful of pilot users to tens of thousands of pull requests reviewed each week, with a 20x jump in ARR through 2024

• Recently launched Diamond, an AI reviewer, which reviews “tens of thousands” of code changes (known as pull requests) every week

👫 Founders

💼 Opportunities

• Software engineer

• Senior frontend engineer

• Senior design engineer

• Staff product designer

• Account executive

• Solutions engineer

• Product marketing lead

• Senior software engineer

🔮 Our Analysis

Code review has become a bigger bottleneck in modern software teams as AI tools like Windsurf and Cursor accelerate code generation. Graphite focuses on the “outer loop” of engineering, where the real delay often lies: reviewing, merging, and shipping code. Its stacked pull request model breaks down large pull requests into smaller, manageable chunks that are easier to review and merge, and Diamond, its AI code review agent, adds another layer of speed by surfacing actionable, high-signal feedback automatically.

One of Graphite’s core advantages is its seamless fit within existing workflows. Graphite integrates directly with GitHub, supports VS Code, and mirrors how developers already work, reducing friction. Teams don’t need to change version control platforms or CI/CD setups to benefit. Graphite simply layers on top. 

Graphite's AI focus also helps it stand out. Diamond claims a sub‑3% false positive rate and comes with a contextual understanding of entire repos, enabling it to act as a reliable co-reviewer. This lets engineering teams handle a higher review load, especially as the speed and volume of generated code continues to rise. This means increased overall code and feature throughput.

Looking ahead, Graphite’s main challenge will be competing with GitHub’s own evolving suite of AI and review tools. Maintaining a high-quality experience at scale, and proving its ROI to larger organizations, will be key as the company pushes further. Still, with a sharp focus, growing traction, and strong backing, Graphite has what it takes to become a staple in the modern developer’s toolkit.

📚 Further reading

• Anthropic-backed AI-powered code review platform Graphite raises cash [TechCrunch]

• Graphite - The end-to-end developer platform [Graphite]

• Graphite debuts Diamond AI code reviewer, insists 'AI will never replace human code review' [DEVCLASS]

Written by Arjun Raj Loomba

🎤 Founder Q&A

This Q&A has been reconstructed from interview notes with Merrill Lutsky, CEO and Co-Founder of Graphite.

At the time, we were building a feedback tool for hotels and restaurants. We ended up selling that to another company building SaaS tools for hotels. This helped me get experience with starting a company, building a team, raising money, going door to door, talking to hotels in SF, and building out a product.

When I got back to Harvard, I started immersing myself in the startup space on campus, and I got to know (Reimers) and (Foster) through that. I also worked at Square part time. After we graduated, I moved to NY to join Oscar Health. Tomas joined Facebook in New York. The both of us would hang out in each other’s offices after work, talking about things like S-1 filings, tech strategy, and thinking about what we’d want to build. We formed a close bond from that. Greg was at Google and then (at) Airbnb in SF, working on infrastructure there. Personally, I went back and forth between product and engineering. 

We’d seen what big and small company developer toolchains were, and thought about what the problems were there. We became passionate about developer productivity. The best part about working on such tools as an engineer is that you build something that you get to use every day. With this, we eventually landed on starting Graphite.

With so much attention on AI tools that assist with code generation, like Cursor and Windsurf, what led you to focus on the “outer loop” of software engineering, like code reviews, instead? 

There are a couple of reasons for that. Firstly, we were working on the outer loop before LLMs got really good at writing code. Our goal from the beginning has been to accelerate the process of reviewing, testing, and deploying. In our observation, even before LLMs, that’s (been) the bottleneck for how quickly companies can ship code. 

Larger companies, especially, have so many engineers, and it's not a matter of how quickly you can write the code for a feature, rather how quickly you can get it through the review and deployment processes. 

We’ve been focused on the problem of code review processes for a while, and it just so happened that with the advent of Copilot (and) tools like Windsurf and Cursor that this problem just got ten times worse. Now, engineers can write features faster than before—they have these amazing features to generate code. But the second half of the story is how you test, review, and deploy it, and that’s still the same. That’s a huge opportunity that we are building toward. There’s a lot of focus on generating code, and all of that still needs to be reviewed. We want to apply AI models to help human engineers to handle the volume and the nature of AI-generated code. More so than ever, that’s going to be the rate-limiting factor to how quickly you can ship, especially since the time to generate code gets closer to zero. 

How does Graphite ensure the trustworthiness and safety of its AI code review suggestions, so that developers can confidently rely on them?

We see this tool as augmenting human reviewers, not replacing them. As long as a human reviewer is responsible for what code goes out into production, we don’t see this as an issue. One of our beliefs for the next couple of years is that the activity of software engineers will shift from writing code (and) spending time in your IDE to reviewing code. Humans will kick off tasks to AI agents, and then review and iterate with agents. Until you can have an AI that’s accountable for end user experience, there’s always going to be a human in the loop. 

To handle the volume and speed of AI-generated code that comes out now, you need AI tools to navigate and focus the review. This would allow the human to not spend so much time on each pull request and focus on the higher level. This is where Graphite’s AI code review tool, Diamond, is so good. It can find bugs, smaller issues or nits (nitpicks), and style guideline inconsistencies. Everything you’d be looking for a first pass on a pull request. While the pull request reaches you, the author has solved all the basic issues, and you can review the main parts.

Being high signal and high trust is important for engineers to gain real value out of Diamond. We’ve incorporated an upvote and downvote system. We’ve been optimizing our evals, how we’ve been architecting our prompts. Every time Diamond leaves a comment, it’s at least as trustworthy as a human engineer. This ensures a high quality of comments, and increases the rate (at) which Diamond’s suggestions are actually used to change code. The high signal aspect sets Diamond apart from other code review agents.

🥲 That’s all, folks

If you enjoyed today's article, forward this email to a friend!

If you're just seeing this email now, subscribe here.

Made with 💜 by the Unicorner team